US and European mutual fund performance and a
comparison with Exchange Traded Funds
Master Thesis Roeland de Bruin
MSc Business Economics, Finance Track Master Thesis
Supervisor: P. Versijp July 2015
Statement of Originality
This document is written by Roeland de Bruin who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in
creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of content
1. INTRODUCTION ... 5
2. LITERATURE REVIEW ... 11
2.1MUTUAL FUND PERFORMANCE ... 11
2.2MUTUAL FUND PERFORMANCE AND THE RESIDUAL BOOTSTRAP METHODOLOGY ... 13
2.3SURVIVORSHIP BIAS ... 14
2.4MUTUAL FUND PERFORMANCE DURING THE 2008GLOBAL FINANCIAL CRISIS ... 14
2.5ETFS ... 15 3. HYPOTHESES ... 16 4. SAMPLE ... 18 5. DATA ... 19 6. METHODOLOGY ... 27 6.2RESIDUAL BOOTSTRAP ... 28 6.3BOOTSTRAP PROCEDURE ... 28 7. EMPIRICAL RESULTS ... 30
7.1OLS REGRESSION RESULTS ... 30
7.2BOOTSTRAP ESTIMATIONS ... 38
7.3COMPARISON BETWEEN MUTUAL FUND AND ETF SAMPLES ... 41
8. ROBUSTNESS CHECKS ... 43
9. CONCLUSIONS AND DISCUSSION ... 46
9.1MUTUAL FUND PERFORMANCE ... 46
9.2ROBUSTNESS CHECKS ... 47
9.3COMPARISON WITH ETFS ... 48
REFERENCE LIST ... 50
APPENDIX ... 52
Abstract
This research investigates mutual fund performance in the US and Europe between 2002 and 2014, with emphasis on the 2008 Global Financial Crisis. Furthermore the performance of the
mutual funds is compared with an Exchange Traded Fund (ETF) sample, a not-actively managed investment product. A residual bootstrap procedure is used to overcome
non-normalities in mutual fund returns. The results show that, in contrast to most literature covering earlier periods, US mutual funds slightly outperform the Fama & French benchmark.
54.27% of the US mutual funds have outperformed the Fama & French benchmark. This outperformance has increased during the crisis period. Of the European mutual funds, 93.77%
have underperformed the Fama French benchmark, and the underperformance is even worse during the crisis. I also compared the performance of the mutual fund samples with the Russell 3000 index and the MSCI Europe index. The relative performance slightly changes,
but the US mutual funds still outperform the market and European mutual funds still underperform. For both periods in both regions, the ETFs performed better than the mutual
1. Introduction
This research aims to examine whether European and US investors would have been better off buying an Exchange Traded Fund (ETF) than a mutual fund (which invests in the US of European equities) between 2002 and 2014, with emphasis on the 2008 global financial crisis. Where mutual funds are actively managed and therefore charge high management fees (Fama and French, 2010), ETFs are genuinely significantly cheaper, because those are not actively managed but simply track an index. The question is whether mutual fund managers have enough skill to compensate their relatively high management fees and whether they are able to outperform passive ETFs. On the other hand, ETFs don’t have the freedom of action, where mutual funds have. So the mutual funds might have minimized their losses during the crisis by reducing their holdings temporarily, where ETFs fall with the market. This difference between mutual funds and ETFs makes it an interesting period to investigate.
According to the 2014 Investment Company Fact Book, the total assets under management in the mutual fund industry were $30 trillion at the end of 2013. The industry has rapidly grown over the last 15 years, as shown in figure 1 below. The large size of the mutual fund industry might be justified by the large amount of studies published on mutual funds. Popular topics researches on mutual funds are performance and possible persistence in performance, possible market timing, asset allocation and security selection skills of mutual fund managers and the survivorship bias in mutual fund performance. Most of the researches regarding mutual funds focus on the US market, some studies on European mutual funds (Otten and Schweitzer, 2002; Otten and Bams, 2002) and some studies on national fund markets (Khorana, Savaes and Tufano, 2005). The results regarding US mutual fund performance are contradictory. Most of the papers (e.g. Fama and French, 2010; Malkiel, 1995; Jensen 1968) found that US mutual funds underperform their benchmarks after fees. The main cause for this underperformance is that mutual funds are actively managed and therefore charge high management fees. However, there are some papers who found that mutual fund are able to outperform their benchmark and therefore add value to investors (e.g. Cremers and Petajisto, 2009); Wermers, 2000).
On European mutual funds, only Otten and Bams (2002) and Otten and Schweitzer (2002) investigated the performance of European mutual funds. Otten and Bams (2002) found that most European mutual funds are able to outperform their benchmarks and Otten and Schweitzer (2002) found that European mutual funds perform better than US mutual funds. European mutual funds might perform better than US mutual funds, since the expense ratio of European funds is lower. Furthermore, Otten and Bams (2002) conclude that it might be more
difficult to outperform the market in the US since the mutual fund industry is relatively larger. When the mutual fund industry is larger relative to the total market (compared to Europe), it’s more difficult to outperform the market as a group (Otten and Bams, 2002). This research examines US and European mutual fund performance using a bootstrap methodology on a survivorship bias-free sample. Not yet a bootstrap procedure have been applied on a European mutual fund sample. Furthermore, the performance of the mutual funds will be compared with a sample of ETFs.
While mutual funds already exist for a very long time, an Exchange Trade Fund (ETF) is a relatively new long-term investment product; the first ETF became available in 1993 (Poterba and Shoven, 2002). ETFs became very popular; the ETF industry has grown rapidly since their inception, as shown in figure 1 and 2. In the first 10 years, the growth mainly stemmed from a relatively small number of ETFs, which became larger. Later on, from up 2006, the number of ETFs has grown rapidly. This is also shown in figure 2. Although the mutual fund industry is still 10 times larger than the ETF industry, the growth of the ETF industry has been much higher between 2002 and 2013.
Figure 1: AUM in mutual fund industry
Total assets under management in the mutual fund industry and the ETF industry, 1996-2013. Mutual fund data include only mutual funds that report statistical information to the Investment Company Institute.
Source: Investment Company institute (2014)
0 200 400 600 800 1000 1200 1400 1600 1800 0 2.000 4.000 6.000 8.000 10.000 12.000 14.000 16.000 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 201 1 2012 2013 ETF as se ts , b il li on $ M u tu al fu n d as se ts , b il li on $ Mutual Funds ETFs
Figure 2: number of ETFs per year
The figure presents the total number of ETFs in the US between 2002 and 2013.
Source: Investment Company institute (2014)
As said, mutual funds are actively managed and therefore charge high management fees (Fama and French, 2010). Otten and Schweitzer (2002) found that in the US and in Europe the average management fee is respectively 1.4% and 1.2%. In Europe, they found that especially Italian and Spanish mutual funds charge above average management fees, namely 2%. Furthermore, Luo (2002) found that an investor holding a fund from the beginning of 1995 to the end of 1997, paid an average annual fee of 1.76%. So, mutual fund managers have to pick winner stocks consistently to compensate for these costs. ETFs simply track an index and are therefore cheaper for investors. This research investigates whether mutual fund managers have enough skill to outperform the market and ETFs after fees.
Most of the researches regarding mutual fund performance have been done in the previous century and in the early 2000s. This research investigates the performance of mutual
funds during a more recent period.One possible reason for different results in a more recent
period is the rise of ETFs. Cremers et al (2015) found that explicit indexing improves competition in the mutual fund industry. Since the ETF industry is growing, the increasing competition will stimulate mutual funds to charge lower fees. These lower fees should improve fund performance. Another reason for different results could be that mutual funds might have changed their investment strategies during and after the Dot-com bubble around
0 200 400 600 800 1000 1200 1400 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 N u m b er o f E T Fs p er y ea r
2000 and the 2008 global financial crisis. Where mutual funds can adjust their holdings and change their investment strategies, ETFs don’t have the freedom of action.
Besides investigating a more recent period, it’s also interesting to examine the performance of mutual funds during the most recent global financial crisis (2007-2008), mainly because Ben-David, Franzoni and Moussawi (2012) found that mutual funds reduced their holdings throughout 2008. As said, ETFs don’t have the freedom of action and thus not the possibility to adjust their holding. The active phase of the financial crisis started in the first week of august, when quantitative hedge funds suffered unprecedented losses for no obvious reason (the Quant Meltdown). On 9 August 2007, BNP Paribas terminated withdrawals from three hedge funds, citing “a complete evaporation of liquidity”. Next, in December 2007 the subprime mortgage crisis started in the US, when the bubble in the US housing market bursted. In the US, the crisis ended with the increasing stock prices from up March 2009 (Ben-David, Franzoni and Moussawi, 2012), as shown in figure 3 on the next page.
The 2007-2008 financial crisis and the Subprime mortgage crisis were the main triggers for the European debt crisis. From up the end of 2009, some European states were unable to repay or refinance their government debt, especially Greece. This resulted in fears that some European countries could be forced to leave the European Monetary Union. This uncertainty influenced European stock prices negatively, as shown in figure 4. In 2012, the S&P500 had almost reached her pre-crisis level again, where the European stocks were still close to the low of 2009. So, I consider the crisis period longer for Europe than for the US. In June 2012, the pro-austerity party new Democracy won the elections in Greece, allaying fears the country was about to leave the Eurozone. After the election the European stock prices rose, as shown in figure 4: the chart of the Euro STOXX 50 index (an index with the 50th largest European companies). In my research, the Greek elections of June 2012 (and the rising stock prices after the election) are seen as the end of the crisis period in Europe. As said, ETFs don’t have the freedom of action and are thus not able to adjust their holdings during these crises. A possible reduction of holdings of mutual funds might have resulted in an outperformance for the mutual funds during this specific period.
Figure 3: Price chart S&P500 index, 2007-2014.
This graph presents the price development of the S&P500 index from 2007 through 2014. It confirms the crisis period in the US between August 2007 and February 2009.
Figure 4: Price chart Euro STOXX 50 index, 2007-2014
This graph presents the price development of the Euro STOXX 50 index from 2007 through 2014. The graph confirms the crisis period between August 2007 and May 2012.
The research question of this paper is: do US and European equity mutual funds outperform the market and Exchange Traded Funds?
My results indicate that US mutual funds slightly outperform both the Fama & French benchmark and the Russell 3000 index between 2002 and 2014. The outperformance increases during the crisis. However, the US mutual funds underperformed the ETF sample during both periods. The European mutual funds underperformed both the Fama & French benchmark and the MSCI Europe index between 2002 and 2014. This underperformance increases during the European crisis period. The European mutual funds are thereby not able to outperform the ETF sample.
The research continues with a literature review, in which I’ll provide a background for the research topic. Then the hypotheses will be formed. Next, I’ll explain how I compose my sample and from where I obtain my data. Next, in the methodology, I’ll explain how I’m
going to test mutual fund performance and how I will compare this performance with ETFs. Next I’ll present the results and give them economic meaning. Lastly I will further interpret my results in the conclusion and discussion. The appendix can be found at the end of my paper.
2. Literature review
2.1 Mutual fund performance
As said in the introduction, lots of research has been done regarding (US) mutual fund performance. The results of these papers are contradictory; most of the papers (Malkiel (1995); Cuthbertson, Nitzsche and O’Sullivan (2010); Fama and French 2010) found that US mutual fund on aggregate underperform their benchmark. However, Kosowski et al (2006) found evidence that a sizeable minority of mutual fund managers pick stocks well enough to persistently outperform their benchmarks after costs and Cremers and Petajisto (2009) found that funds with highest active share significantly outperform their benchmark, both before and after fees. Below I first summarize some of the relevant papers regarding mutual fund performance.
Otten and Bams (2002) present an overview of the European mutual fund industry and examine their performance, using the Carhart (1997) four-factor model on a sample of 506 mutual funds. Thereby they investigate whether the performance is persistent. They construct a mutual fund database of the five most important mutual fund countries, which together covers over 85% of total assets in European mutual funds. They found that in 4 out of the 5 countries the mutual funds, on aggregate, outperform their benchmark. In the United Kingdom they found persistence in mean returns. They conclude that their results deviate from most of the studies on US mutual funds. This can be explained by the fact that the US mutual fund industry is larger relative to the market, and therefore it’s more difficult to outperform the market as a group. The findings of Otten and Bams (2002) regarding European mutual fund performance are important in forming my hypothesis later on.
In line with this research, Otten and Schweitzer (2002) compare the development and performance of the European mutual fund industry with the US mutual fund industry. They found that the US mutual fund industry is much larger, but that the number of funds in Europe is higher. The European funds are thus smaller. Furthermore they use a traditional structure-conduct-performance (SCP) paradigm to study the effects of the structure and conduct on the performance in the mutual fund industry. The SCP is a framework developed in organizational literature that focuses on product and production efficiency. They combine the SCP paradigm with asset pricing model in their performance measure. Besides performance they also compare the expense ratios of mutual funds in the different continents. They found that the expense ratios are on average slightly lower in Europe (1.2% in Europe compared to 1.4% in the US), with in Europe both higher minima and lower maxima. Finally they found
that the European mutual funds perform better then the US mutual funds, and that in both the US and in Europe small cap funds outperform their benchmarks and all other funds. The lower fees in Europe seem to be an important driver why European funds performed better than US funds. The research of Otten and Schweitzer (2002) is important in forming the hypotheses for this research and making the comparison between US and European mutual funds.
Another paper which is important in forming the hypotheses is the research of Cuthbertson, Nitzsche and O’ Sullivan (2010). They evaluate academic research of mutual fund performance in the US and the UK. They state that the evidence suggest that there are around 2-5% of top performing UK and US equity mutual funds which genuinely outperform their benchmark (after fees). 20-40% of the US and UK funds performed genuinely poor relative to their benchmarks. Key drivers for the relatively bad performance of the mutual funds are fees, expenses and high turnover. Furthermore they found little evidence of market timing. They do found evidence on picking winner stocks that suggest the past winners persist. However, due to transaction costs and fund fees, the economic gain of these winner funds may be marginal. In recent research, however, they found that when funds are using more sophisticated sorting rules (Bayesian approached), funds possibly earn large gains from picking winner stocks. Another interesting finding is that evidence clearly supports the view that past loser funds remain losers. Cuthbertson, Nitzsche and O’ Sullivan advise investors to hold low cost index funds and avoid holding past active losers, since past loser funds remain losers.
Lastly Wermers (2000) examines mutual fund performance, and decomposes the performance into stock-picking talent, style, transaction costs and expenses. He starts with asking the question whether mutual fund managers who actively trade stocks add value. He states that academics have debated this question since the paper of Jensen (1968), but now the majority of papers conclude that actively managed funds on average underperform passively managed counterparts. He finds that funds hold stocks that outperform the market by 1.3% per year, but after fees the funds underperform the market by 1%. Of the 2.3% difference between the gross and net return, 0.7% is due to underperformance of non-stock holdings and the other 1.6% is split almost evenly between expenses and transaction costs. Thus, considering only their stock holdings, Wermers (2000) concludes that mutual fund managers hold stock that beat the market portfolio by almost enough to cover their expenses and transaction costs. Thereby, high-turnover funds beat the Vanguard Index 500 fund on a net return basis. Wermers’ evidence thus supports the value of active mutual fund management.
2.2 Mutual fund performance and the residual bootstrap methodology
Fama and French (2010) examine mutual fund performance in the US. They examine whether US mutual funds outperformed their benchmarks between 1984 and 2006, and next they try to distinguish luck from skill in mutual fund performance. They use both the Fama and French three-factor model (1993) and the Carhart (1997) four-factor model. They found that mutual funds, on aggregate, underperform the CAPM, thee-factor and four-factor benchmarks by about the amount they charge fees to the investors (after bootstrap simulations). Thereby they found that few funds have enough skills to cover the costs. This Carhart (1997) four-factor model is most commonly used in researches regarding mutual fund performance. The methodology of this research is partly based on the methodology of Fama and French (2010).
Furthermore on US mutual funds, Kosowski et al (2006) investigate the performance of the US open-end, domestic-equity mutual fund industry between 1975-2002, using a statistical technique called bootstrap. They state that a bootstrap approach is necessary, because 1) the cross-section of mutual fund alphas has a complex, non-normal distribution
(heterogeneous risk-taking by funds leads to fatter tails than in a normal distribution) and
because of 2) non-normalities in individual fund alpha distributions (thus normality may be a poor approximation in practice). A bootstrap estimation can substantially improve on this approximation. They use the Carhart (1997) four-factor model to compute OLS estimators. Next they use a residual bootstrap technique to obtain time-series of pseudo monthly excess returns for each fund. They found that a sizeable minority of mutual fund managers pick stock well enough to persistently outperform their benchmarks after costs. Furthermore, their evidence indicates that a bootstrap procedure is necessary for future research on mutual fund performance. The bootstrap methodology of Kosowski et al (2006) can also be applied to the mutual fund and ETF samples of this research.
Blake et al (2014) compare the bootstrap methods of Kosowski et al (2006) and Fama and French (2010), using a data set on UK equity mutual funds. To be able to make a good comparison, they use the same dataset for both bootstrap methodologies. They use the Carhart (1997) four-factor model to compute OLS estimators. Next, they perform both bootstrap methodologies. They found that the average equity mutual fund manager in the UK is unable to outperform the market, using either stock selection or market timing. Furthermore, 95% of fund managers on the basis of the Kosowski et al (2006) bootstrap and almost all fund managers on the basis of the Fama and French (2010) bootstrap failed to outperform the market after bootstrap simulations. Thus, the majority of the mutual fund managers are not simply unlucky, but they are on average unskilled. The mutual fund managers who are able to
outperform the market, extract the excess returns for themselves via fees, leaving no outperformance for investors. The literature suggests that a bootstrap procedure is necessary for my research. Blake et al (2014) give a helpful insight of the bootstrap procedures.
2.3 Survivorship bias
Malkiel (1995) investigates the performance of mutual funds between 1971 and 1991 and investigate the magnitude of the survivorship bias. He found that papers that systematically exclude not-surviving mutual funds significantly overestimate the returns received by mutual fund investors. Using a sample with only mutual funds that exist continuously in the sample period 1971-1991, he found an insignificant negative alpha. His results imply that even surviving funds do not generate excess returns after expenses. He found that when returns of all funds are investigated, mutual funds are tended to underperform the market. Malkiel (1995) concludes that investors would be considerably better off by purchasing a low-cost index fund, rather than trying to find an active manager who can persistently perform well. In line with Malkiel (1995), this research also investigates whether investors are better of buying a passive investment (the ETF) instead of an actively managed mutual fund.
In line with the research of Malkiel (1995), Elton, Gruber and Blake (1996) examine mutual fund performance and the effect of survivorship bias on the outcomes of mutual fund researches. They state that mutual fund attrition can create a biased sample, because the funds that will be dropped are likely to be bad performing funds. When using the remaining sample, the estimates will overstate the actual performance of the mutual funds industry. In the majority of cases, the bad performing funds don’t disappear but are merged into another fund. The record of the funds’ poor performance is most likely deleted from the databases. They examine the impact of the survivorship bias by calculating the performance of mutual funds using two samples, one free of survivorship bias and one sample with survivorship bias. They found a difference in their estimations. The difference becomes larger when the number of years in the study increases. Hence, given the results of Malkiel (1995) and Elton, Gruber and Blake (1996), it is important to correct for survivorship bias in this research.
2.4 Mutual fund performance during the 2008 Global Financial Crisis
As said in the introduction, this research also examines mutual fund and ETF performance during the 2008 Global Financial Crisis. I’ll use the research of Ben-David, Franzoni and Moussawi (2012) to gain more insight in trading activities of mutual funds during the 2008 global financial crisis. Ben-David, Franzoni and Moussawi (2012) examine the differences in trading activity of hedge funds during the 2008 global financial crisis, and they compare these
differences with the differences in trading activity of mutual funds, other institutional investors and non-institutional investors. Among others they observed that mutual funds reduced their holdings throughout 2008. This finding leads to the question whether the mutual funds (with reduced holdings) outperformed their benchmarks during 2008, and thus whether the reduction of holdings has paid off for the mutual funds.
2.5 ETFs
In this research we compare mutual funds with ETFs. Below we summarize some papers regarding ETFs’ characteristics and performance. Since ETFs are relatively new investment products, relatively less research has been done on this field.
Poterba and Shoven (2002) give an introduction to the operations of ETFs. Furthermore they compare pre-tax and post-tax returns of the largest ETF with the return of the largest equity index mutual fund. They found that both the after-tax and the pre-tax returns of the fund were slightly higher than those of the ETFs.
Agapova (2011) examines implications of the substitutability of conventional index mutual funds and ETFs and she examines the potential explanations of the coexistence of both index mutual funds and ETFs in the market. Agapova (2011) found that conventional funds and ETFs are substitutes, but not perfect substitutes. However, one dollar of ETF flows should be expected to reduce conventional mutual index fund flows by 22 cents, so the existence of ETFs influences the index mutual fund market. She suggests that the coexistence of both instruments can be explained by a clientele effect, that segregates the two vehicles into different market niches.
3. Hypotheses
Given the literature review, the following hypotheses are formed. They will be tested using the research methodology. The hypotheses are numbered. On US mutual funds;
H0: US equity mutual funds underperformed their benchmark after fees. (1)
Although the results on US mutual fund performance are contradictory, the majority of the literature regarding US mutual fund performance found that these funds were on aggregate not able to beat the market. The most important factors that cause this underperformance are high expenses and high transaction costs (due to a relatively high turnover). For the comparison with ETFs, the expectation is that US mutual funds underperform the comparable ETFs, mainly because where mutual funds are actively managed and therefore charge high management fees, ETFs are passive investments and have much lower expense ratios.
H0: US equity mutual funds underperformed comparable ETFs. (2)
Although the papers of Otten and Bams (2002) and Otten and Schweitzer (2002) found contrary results regarding mutual fund performance in Europe, we follow the results of Otten en Bams (2002) in forming the following hypotheses:
H0: European equity mutual funds outperformed their benchmark after fees. (3)
H0: European equity mutual funds outperformed comparable ETFs. (4)
So the expectations of this research are in line with another conclusions of Otten and Schweitzer (2002) that European mutual funds perform better then US mutual funds. This can be explained by the finding of Otten and Schweitzer (2002) that European funds charge, on average, lower management fees. Thereby, the fact that the US mutual fund industry is larger relative to the total market makes it therefore more difficult to outperform the market as a group in the US (Otten and Schweitzer, 2002).
On the 2008 global financial crisis period, Ben-David, Franzoni and Moussawi (2012) found that US mutual funds reduced their holdings throughout 2008. Given that the stock markets have mostly declined during that year, the following hypotheses are formed.
H0: During the 2008 global financial crisis, US equity mutual funds outperformed their
benchmark after fees.
(5)
H0: During the 2008 global financial crisis, US equity mutual funds outperformed
comparable ETFs.
(6)
There is no literature regarding European mutual fund performance during the 2008 global financial crisis and the Eurocrisis, so the hypotheses are the same as the hypotheses for the 2002-2014-sample period. Thus, I expect that European mutual funds outperform their benchmarks and ETFs.
Note that the statistical zero-hypotheses are no under- of outperformance of mutual funds and no difference between the performance of mutual funds and ETFs.
4. Sample
For both US mutual fund sample, only mutual funds that invest in general (small-, mid-, of large-cap) US equity will be incorporated. Logically bond funds, mixed-assets funds and international or global funds are excluded from the sample. Thereby, also the funds that invest in specific US states, specific sectors or in US real estate are excluded from the sample. On the European mutual fund sample, Solely mutual funds that invest in general European equity are used. Again, specific sector funds are excluded from the sample. In both mutual fund samples (passive) index funds are excluded.
The mutual fund performance for both US and European mutual funds will be compared with a sample of ETFs. In selecting ETFs, physical replication ETFs and synthetic ETFs have to be distinguished. In a physical ETF, investors money is directly invested in the index’ stocks in order to replicate the index return. In synthetic ETFs on the other hand, the issuer enters into a total return swap with a financial institution, which promises to deliver the performance of the index to the ETF (Hurlin et all, 2014). When using synthetic ETFs, the default risk to the counterparty should also be incorporated. For this research we only use physical replication ETFs that track a broad equity index in the US of Europe. For the US, there are quite much ETFs available with data for the entire period. An overview of all ETFs that will be used can be found in table 1. In panel A the US ETF sample is presented. One of the ETFs that will be used is the Vanguard Total Stock Market. This ETF seeks to track the CRSP US Total Stock Market Index and covers 99,5% of the total US stock market. Furthermore, the Vanguard S&P500 will be used. The S&P500 index contains the 500 largest (based on market capitalization) US firms. Panel B presents the European ETF sample. In this sample only large-cap ETFs will be used, since there are no other (broad) ETFs available with data for the entire period. One of the four ETFs included in the sample is the iShares STOXX Europe 50 UCITS ETF. This ETF tracks the STOXX Europe 50 Index, an index with 50 European companies with the highest market capitalization.
Different ETF samples are created for the crisis period. Since there are more ETFs available in both the US and Europe, both samples are larger than those for the 2002-2014 period. Since these samples consist of many ETFs, there is no table presented in the appendix. These samples consist of small, mid and large cap and broad ETFs from the three largest ETF suppliers: iShares (BlackRock), SPDR (State Street Global Advisors) and Vanguard.
5. Data
Data on US mutual funds can be generated from the CRSP survivorship-bias free US mutual fund database (these are net returns). As said, only US mutual funds that invest in general US equity will be incorporated in the sample. Data on European mutual funds is obtained from Thomson Reuters Datastream. Again, solely mutual funds that invest in general European equity (stocks) will be used. Funds that are listed in France, Germany, the United Kingdom, the Netherlands, Spain and Italy are used. In all countries but the UK funds that are listed in Euros will be used. In the UK, most funds are listed in British Pounds (GBP). Since I investigate mutual fund performance from the perspective of a EuroZone investor, I have to convert the returns of the funds listed in the UK from GBP to Euro. Following Solnik (1974), returns are converted using the historical prevailing sport exchange rates (EUR/GBP). Eight funds were deleted from the sample because of unrealistic data, which are caused by pricing errors in Datastream. Furthermore it’s important to be aware that management fees and other costs are not incorporated in the funds returns from Thomson Reuters Datastream. However, the Total Expense Ratio (TER) of the mutual funds can be downloaded separately from Datastream. But for most funds, there are some dates for which no TER is reported. To calculate the TER for these missing dates we take the average TER of the observations that are available. Next we discount the annual TER to monthly costs and subtract these from the monthly return. These net returns will be used. For both mutual funds and ETFs monthly data will be used.
Data on the ETFs that are used can be found at Thomson Reuters Datastream and the website of the issuers. For the US, only mutual funds and ETF that are listed in USD are used. This implies that this research compares US mutual funds and ETFs from the perspective of a US investor. For non-US investors, exchange risk need to be incorporated in the returns. For the European sample, only mutual funds and ETFs that are listed in EUR will be incorporated in the samples. Furthermore, for the EU ETF sample, we only use ETFs with the Undertakings for Collective Investment in Transferable Securities (UCITS) label. This means that the investment product is established and authorized in one of the EU Member States and designed for EU retail investors. So, this research compares EU mutual funds and ETFs from the perspective of a EuroZone investor. For investors outside the EuroZone, exchange risk has to be incorporated. Dividends are incorporated in the ETF returns, and the returns are corrected for their Total Expense ratios (TER).
To control for survivorship bias in mutual fund performance, we use all mutual funds with data during the period. Malkiel (1995) and Elton, Gruber and Blake (1996) found that
when using a sample with solely funds with data for the entire period, the results would be overestimated, since in that case merged and liquidated funds are excluded from the sample. These fund are often merged or liquidated because of bad performance. So when using only funds with data for the entire period, a group of mostly bad performing funds will be excluded from the sample. Figure 5 illustrates that there have been a significant number of funds merged of liquidated between 2003 and 2013, so it is important to take the survivorship bias into account.
Figure 5: Number of US mutual leaving and entering the mutual fund industry
This figure presents the total number of opened, merged and liquidated mutual funds in the US between 2003 and 2013. These numbers emphasize the importance of taking the survivorship bias into account.
Source: Investment Company institute (2014)
I will investigate the mutual fund and ETF performance for the period 2002-2014 and during the 2008 global financial crisis. For the 2008 global financial crisis, a specific sample will be created, again with data of all equity mutual funds that are active during the period. For the crisis period in the US, we use all data between August 2007 (Quant meltdown) and February 2009 (though stock prices from up March 2009), following Ben-David, Franzoni and Moussawi (2012). In Europe, the 2008 global financial crisis was followed by the European debt crisis. So for Europe, the crisis period is between August 2007 and May 2012, since the Greek elections in June 2012 allayed the fears of a Grexit (Greece leaving the Eurozone). These periods are supported by the stock prices, as shown in figure 3 and 4. Panel A of table 2 presents the descriptive statistics for the US mutual fund sample during the 2002-2014 period.
The total number of funds in the sample is 5,731 and there are 627,929 observations. The average monthly return (RET) of the mutual fund sample is 0.55%, with a minimum of -50.72% and a maximum of 38.62%. From the other variables, the market return (MKTRET) has the highest average return (0.47%). The returns of SMB, HML and MOM are 0.34%, 0.27% and 0.15% respectively. Panel B of table 1 presents the descriptive statistics of the US ETF sample. The sample consists of 31 ETFs (a list can be found in table 1), which corresponds with 4,836 observations. The average return (RET) of the ETF sample is 0.75%, with a minimum of -24.4% and a maximum of 19.9%. The average return is higher for the ETF sample, and the minimum and maximum are less extreme. The average market return (MKTRET) is also higher for the ETF sample, 0.58%. The average returns of the other variables are lower: 0.28% for SMB, 0.20% for HML and lastly 0.10% for MOM.
Table 2 - Descriptive statistics for the US mutual and ETF sample
This table presents the descriptive statistics for the US mutual fund and the US ETF sample. Panel A presents the descriptive statistics for the US mutual fund sample. This sample contains of 5,731 funds and 627,929 observations. Panel B presents the descriptive statistics of the US ETF sample. This sample consists of 31 ETFs, and it contains 4,836 observations. The variable RET are the monthly returns of the funds/ETFs. MKTRET is the Fama-French market excess return factor, defined as the market return minus the risk-free rate. The market return is the value-weight return of all CRSP firms incorporated in the US and listed on the NYSE, AMEX of NASDAQ and the risk-free rate is the one-month T-Bill rate. SMB is small minus big and HML is high minus low. The construction of SMB and HML follows Fama and French (1993). The momentum factor (MOM) is defined by Carhart (1997) as the momentum factor, and will be used to capture the one-year momentum anomaly of the Fama French (1993) model. The period is 2002 through 2014.
Panel A - US mutual fund sample # Funds # of obs Min Mean Max
Standard
Deviation Skewness Kurtosis RET 5,731 627,929 -0.507 0.00555 0.386 0.0490 -0.624 5.003 MKTRET 627,929 -0.172 0.00465 0.114 0.0434 -0.661 4.308
SMB 627,929 -0.052 0.00336 0.059 0.0245 0.115 2.482
HML 627,929 -0.099 0.00270 0.076 0.0236 -0.517 5.133
MOM 627,929 -0.347 0.00153 0.125 0.0499 -2.394 16.721 Panel B - US ETF sample
RET 31 4,836 -0.244 0.00753 0.198 0.0480 -0.673 5.006
MKTRET 4,836 -0.172 0.00578 0.114 0.0433 -0.687 4.439
SMB 4,836 -0.0519 0.00283 0.0585 0.0240 0.125 2.520
HML 4,836 -0.0987 0.00196 0.0757 0.0234 -0.440 5.249
In table 3 on the next page the descriptive statistics are presented for the European samples (for the period 2002 through 2012). All returns are monthly. Panel A shows the descriptive statistics for the European mutual fund sample. The sample consists of 1,178 funds and 109,220 observations. In contrast to the US mutual fund sample, the average return of the European funds is negative (-1.3%). It has a minimum of -33.2% and a maximum of 67.5%. The average returns of all other variables are positive. The average return of the momentum factor (MOM) is the highest, 0.87%. The average return of the market return (MKTRET) is a 0.70%. The average return of SMB and HML are lower, 0.18% and 0.28% respectively. Panel B of table 2 shows the descriptive statistics for the EU ETF sample. The sample consists of 4 ETFs (an overview can be found in table 1). The average return of the ETF sample is 0.34%, which is higher than the average return of the mutual fund sample (-1.3%). For the ETF sample the minimum return is -24.1% and the maximum is 22.1%. These values are again less extreme than the extreme values of the mutual fund sample. The average return for the momentum factor (MOM) and the market return (MKTRET) are also relatively high, 0.90% and 0.69% respectively. Lastly the average return of the other variables SMB and HML are 0.19% and 0.30% respectively.
Table 3 – Descriptive statistics for the EU mutual fund and ETF sample
This table presents the descriptive statistics for the EU mutual fund and the EU ETF sample. Panel A presents the descriptive statistics for the US mutual fund sample. This sample contains of 1,178 funds and182,590 observations. Panel B presents the descriptive statistics of the US ETF sample. This sample consists of 4 ETFs, and it contains 624 observations. The variable RET are the monthly returns of the funds/ETFs. MKTRET is the Fama-French market excess return factor. SMB is small minus big and HML is high minus low. The construction of SMB and HML follows Fama and French (1993). The momentum factor (MOM) is defined by Carhart (1997) as the momentum factor, and will be used to capture the one-year momentum anomaly of the Fama French (1993) model. The European Fama French factors include stocks from Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, the Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, and the United Kingdom. The period is 2002 through 2014.
Panel A - EU mutual fund sample # ETFs # of obs Min Mean Max
Standard
Deviation Skewness Kurtosis RET 1,178 109,220 -0.332 -0.01300 0.675 0.0457 -0.578 6.895 MKTRET 109,220 -0.221 0.00703 0.138 0.0564 -0.668 4.547
SMB 109,220 -0.069 0.00176 0.049 0.0195 -0.329 3.618
HML 109,220 -0.046 0.00284 0.075 0.0217 0.189 3.311
MOM 109,220 -0.260 0.00867 0.138 0.043 -1.751 13.01 Panel B - EU ETF sample
RET 4 624 -0.241 0.00343 0.221 0.0514 -0.513 5.246
MKTRET 624 -0.221 0.00668 0.138 0.0565 -0.651 4.512
SMB 624 -0.069 0.00194 0.049 0.0196 -0.331 3.585
HML 624 -0.046 0.00296 0.075 0.0217 0.176 3.304
MOM 624 -0.260 0.00901 0.138 0.0431 -1.746 12.93
Next the descriptive statistics are presented for the crisis period (August 2007-February 2009 for the US and August 2007-May 2015 for the EU sample). Table 4 on the next page presents the descriptive statistics for the US sample, panel A for the mutual fund sample and panel B for the ETF sample. As shown in panel A, the sample contains 4,101 funds and 75,195 observations. The average return of the mutual funds during the crisis period is -3.08%. This is smaller than the average return during the 2002-2014 period (0.55%), which seems logical since the financial crisis went hand in hand with declining stock price. The minimum return is -50.7% and the maximum 19.7%. The maximum value is low with respect to the maximum for the US mutual fund sample during the 2002-2014 period. The market return (MKTRET) and HML also have negative average return, -3.20% and -1.12% respectively. The average return of the other Fama French factors SMB and MOM are 0.023% and 2.04% respectively.
Panel B shows the descriptive statistics for the US ETF sample during the crisis period. The sample consists of 57 ETFs, with corresponds to 1,140 observations. The average return is -3.22%, with a minimum of 24.4% and a maximum of 9.52%. The average return of -3.22% is close to the average return of the mutual fund sample, and the extreme values here are less extreme than those of the US mutual fund crisis sample. The average returns of the Fama French factors are comparable to those of the mutual fund sample: -3.27% for MKTRET, 0.031 for SMB, -1.16% for HML and 2.03% for MOM.
Table 4 - Descriptive statistics for the US mutual and ETF sample during the crisis period This table presents the descriptive statistics for the US mutual fund and the US ETF sample during the US crisis period (August 2007 till February 2009). Panel A presents the descriptive statistics for the US mutual fund sample. This sample contains of 4,101 funds and 75,195 observations. Panel B presents the descriptive statistics of the US ETF sample. This sample consists of 57 ETFs, and it contains 1,140 observations. The variable RET are the monthly returns of the funds/ETFs. MKTRET is the Fama-French market excess return factor, defined as the market return minus the risk-free rate. The market return is the value-weight return of all CRSP firms incorporated in the US and listed on the NYSE, AMEX of NASDAQ and the risk-free rate is the one-month T-Bill rate. SMB is small minus big and HML is high minus low. The construction of SMB and HML follows Fama and French (1993). The momentum factor (MOM) is defined by Carhart (1997) as the momentum factor, and will be used to capture the one-year momentum anomaly of the Fama French (1993) model.
Panel A - US mutual fund sample during crisis # Funds # of obs Min Mean Max
Standard
Deviation Skewness Kurtosis RET 4,101 75,195 -0.507 -0.0308 0.197 0.0648 -0.738 3.378 MKTRET 75,195 -0.172 -0.0320 0.0460 0.0559 -0.688 2.776 SMB 75,195 -0.0364 0.000223 0.0399 0.0217 0.415 2.328 HML 75,195 -0.0986 -0.0112 0.0441 0.0332 -0.729 3.749 MOM 75,195 -0.0783 0.0204 0.125 0.0508 -0.118 2.420
Panel B - US ETF sample during crisis
RET 57 1,140 -0.244 -0.0322 0.0952 0.0655 -0.791 3.237 MKTRET 1,140 -0.172 -0.0327 0.0460 0.0562 -0.668 2.733
SMB 1,140 -0.0364 0.000311 0.0399 0.0218 0.410 2.311
HML 1,140 -0.0987 -0.0116 0.0441 0.0337 -0.723 3.663
Lastly table 5 on the next page presents the descriptive statistics for the EU samples during the crisis (August 2007-May 2012). Panel A shows that the mutual fund sample consists of 995 funds and 46,710 observations. The average return is -2.06%, which is lower than the average return of the EU mutual fund sample during the 2002-2014 period (-1.3%). The difference between the average returns for the two periods is however smaller than for the US sample (0.55% for the normal period against -3.08% for the crisis period). The minimum of the mutual fund return (RET) is 33.2% and the maximum 58.0%. Only the momentum factor (MOM) has a positive average, 0.71%. The average risk premium (MKTRET) is -0.38%. Due to the crisis the average market return is negative, which causes a negative average risk
premium (rm – rf). The other Fama French factors have negative average returns: -0.063% for
SMB and lastly -0.61% for HML. Panel B shows the descriptive statistics for the EU ETF sample during the crisis period. The sample consists of 18 ETFs and 1,044 observations. The average return of the ETF sample is -0.56%. This is higher than the average return of the mutual fund sample during the same period (-2.06%). The minimum return is 22.9% and the maximum 22.7%. The average of the market return (MKT) is -0.46%. The averages for SMB and HML are -0.12% and -0.54% respectively. As in the mutual fund sample, only the momentum factor (MOM) has a positive average, 0.69% here.
Table 5 - Descriptive statistics for the EU mutual and ETF sample during the crisis period This table presents the descriptive statistics for the EU mutual fund and the EU ETF sample during the European crisis period (August 2007 till May 2012). Panel A presents the descriptive statistics for the US mutual fund sample. This sample contains of 995 funds and 57,710 observations. Panel B presents the descriptive statistics of the US ETF sample. This sample consists of 18 ETFs, and it contains 1044 observations. The variable RET are the monthly returns of the funds/ETFs. MKTRET is the Fama-French market excess return factor (rm – rf). The risk-free rate is the 1 month T-Bill. SMB is small
minus big and HML is high minus low. The construction of SMB and HML follows Fama and French (1993). The momentum factor (MOM) is defined by Carhart (1997) as the momentum factor, and will be used to capture the one-year momentum anomaly of the Fama French (1993) model.. The European Fama French factors include stocks from Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, the Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, and the United Kingdom.
Panel A - EU mutual fund sample during crisis # Funds # of obs Min Mean Max
Standard
Deviation Skewness Kurtosis RET 995 46,710 -0.332 -0.0206 0.580 0.0556 -0.317 4.922 MKTRET 46,710 -0.221 -0.00376 0.138 0.0755 -0.278 2.792 SMB 46,710 -0.0470 -0.000633 0.0485 0.0214 0.0857 2.724 HML 46,710 -0.0460 -0.00612 0.0745 0.0257 0.965 3.841 MOM 46,710 -0.260 0.00705 0.0987 0.0532 -2.174 11.447
Panel B - EU ETF sample during crisis
RET 18 1,044 -0.229 -0.00562 0.227 0.0612 -0.135 3.785
MKTRET 1,044 -0.221 -0.00457 0.138 0.0758 0.304 2.868
SMB 1,044 -0.0465 -0.00119 0.0485 0.0219 0.0934 2.700
HML 1,044 -0.0460 -0.00544 0.0745 0.0256 0.940 3.850
6. Methodology
To examine whether the mutual funds in the different continents can outperform their benchmarks and ETFs, I use the Carhart (2007) Four-factor regression model. I chose this model because Carhart (1997) states that his four-factor model significantly improves the three-factor model of Fama and French (1993) (and the CAPM model).
Ri,t = αi + β1,i MKTt + β2,i SMBt + β3,i HMLt + β4,i MOMt + εi,t. (1)
The αi indicates a possible underperformance or outperformance with respect to the market.
Under the statistical null hypothesis of zero abnormal return, the alpha coefficient would be zero. To test whether mutual funds are, on average, able to outperform the market, the average alpha of the entire sample will be tested for significance (Malkiel, 1995). The
variable market excess return (MKTt) is the market return minus the risk free rate. The market
return for the US is the value-weight return of all CRSP firms incorporated in the US and listed on the NYSE, AMEX of NASDAQ, and the risk-free rate is the 1-month T-Bill rate, following Fama and French (2010) and Carhart (2007). The European factors include stocks from Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, the Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, and the United Kingdom. The variable SMBt (small minus big) is the average return on three small portfolios minus the
average return on three big portfolios, the size return. HMLt (high minus low) is the average
return on two value portfolios minus the average return on two growth portfolios, the
value-growth return. The construction of SMBt and HMLt follows Fama and French (1993). MOMt
is defined by Carhart (1997) as the momentum factor, and will be used to capture the one-year momentum anomaly of the Fama French (1993) model. Monthly data on the Fama-French
factors MKTt, SMBt, HMLt and MOMt can be found at the website of Kenneth French, both
for European and US markets.
The Carhart (1997) regression model will be applied for the US and the European mutual fund samples and also for both ETF samples. The constant of the regressions represents the alpha and indicates a possible outperformance compared to the Fama & French benchmark. I’ll compare the mutual fund and ETF samples for both regions by comparing the constant of both regressions, using a generalized Hausman specification test.
6.2 Residual bootstrap
Kosowski et al (2006) reported that a time series of individual mutual fund performance exhibit non-normal distributions. They state several factors causing these non-normalities. First, the residuals of the mutual fund returns are not drawn from a multivariate normal distribution. Furthermore, these residuals might also be correlated, since high-risk funds often hold concentrated portfolios that load on the same industries or individual stocks.
Another cause for the non-normalities is the variation risk-taking between funds.
6.3 Bootstrap procedure
In the methodology I obtained OLS estimates using the Carhart (1997) four-factor model. Next we follow Kosowski et al (2006) and I draw, for each fund i, a random sample with replacement from the funds residuals, which are obtained by the OLS regression, creating a
pseudo time-series of resampled residuals, {ε^b
i,t, t = Ti,0,..,Ti,1}. Next the bootstrap residuals
are randomly reattached to fitted values, to obtain a time-series of pseudo monthly excess returns is created for fund i, imposing a zero abnormal performance (αi = 0; statistical null
hypothesis):
Rbi,t = β^1,i MKTt + β^2,i SMBt + β^3,i HMLt + β^4,i MOMt + ε^bi,t. (2)
Where b is the bth bootstrap and ε^b
i,t is drawn from {ε^bi,t, t = Ti,0,..,Ti,1}. The pseudo
time-series has a zero alpha by construction. However, when running the above regression for a given bootstrap sample, a positive estimated alpha (and t-statistic) may result, since that bootstrap may have drawn an abnormally high number of positive residuals, or, conversely, a negative alpha (and t-statistic) may result if an abnormally high number of negative residuals are drawn. For instance for b=1, the pseudo excess returns will be regressed on the Carhart (1997) four-factor model:
Rbi,t = αi + β^1,i MKTt + β^2,i SMBt + β^3,i HMLt + β^4,i MOMt + ε^i,t. (3)
Repeating the above procedure for each fund, I get a draw from the cross-section of bootstrapped alphas. Repeating this for all bootstrap iterations, b=1,…,1000, a distribution of
the cross-sectional draws of alphas will be created, {α!^!, i = 1,..,N}, with their corresponding
t-statistic, {𝑡!
! ^
histogram. This cross-section results purely from random sampling, while a zero true abnormal performance is imposed. The average bootstrapped alpha will indicate whether the mutual fund have under- or outperformed the market after fees (on average).
7. Empirical results
7.1 OLS regression results
The results of the OLS regressions are presented in table 6 on page 31. The results for the US mutual fund sample are presented in the first column of table 6. The survivorship bias-free sample consists of 5,731 funds and 627,929 observations. The constant of the regression is 0.000122 (0.0122%) and is significant at a 1% significance level. The other coefficients are significant at a 1% significant level as well. The significant, positive constant indicates a small outperformance of the US mutual funds with respect to the Fama & French benchmark. Figure 6 shows the distribution of all individual mutual fund alphas.
Figure 6 – Distribution of individual US mutual funds alpha
From the total sample 54.27% (3,110 out of 5,731 funds) outperformed the Fama & French benchmark. The best performing fund has an alpha of 18.01% and the worst performing fund has an alpha of -12.00%. There are 11 funds that outperformed the benchmark with more than 5%. When I exclude these 11 funds, the constant of the regression remained unchanged. This means that the aggregate outperformance of the US mutual funds is not only caused by a few good performing funds. These result regarding US mutual fund performance are not in line with the findings of Malkiel (1995) and Fama and French (2010), who found that US mutual funds underperform their benchmarks with the amount they charge fees to investors. Thereby
the results are also not in line with the findings of Cuthbertson, Nitzsche and O’Sullivan (2010), who state that only 2-5% of the top performing US mutual funds are able to outperform. But, as said in the literature review, the literature on US mutual funds performance is contrary. My results are in line with the finding of Wermers (2000) who states that actively managed funds add value, and with Cremers and Petajisto (2009) who found that funds with highest active share significantly outperform their benchmark, both before and after fees.
Furthermore, I examine whether there is a pattern visible in the alpha of the US mutual fund sample. Figure 7 presents the alpha of US mutual funds plotted over time. There doesn’t seem to be a pattern in the alpha. The minimum of the graph was at the end of 2008, during which the mutual funds underperformed the Fama & French benchmark by almost 20%. During 2009, 2010 and 2011 there were some peaks during which the mutual funds outperformed the benchmark with more than 10%.
Figure 7 – US mutual fund alpha over time
Table 6 – OLS regression results for all samples (2002-2014 period)
This table presents the OLS estimations of the Carhart (1997) four-factor model for all (MF and ETF) samples. The dependent variable RET are the monthly returns of the funds/ETFs. MKTRET is the Fama-French market excess return factor (rm – rf). For the US, the market return is the value-weight
return of all CRSP firms incorporated in the US and listed on the NYSE, AMEX of NASDAQ. For the European samples, the market return includes returns of stocks from Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, the Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, and the United Kingdom. The risk-free rate is the 1-month T-Bill rate. SMB is small minus big and HML is high minus low. The construction of SMB and HML follows Fama and French (1993). The momentum factor (MOM) is defined by Carhart (1997) as the momentum factor, and will be used to capture the one-year momentum anomaly of the Fama French (1993) model. Robust t-statistics are reported in parentheses. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.
US Mutual Funds US ETF EU Mutual Funds EU ETF
Coefficients Coefficients Coefficients Coefficients (Robust t-statistic) (Robust t-statistic) (Robust t-statistic) (Robust t-statistic) Constant 0.000122*** 0.00100*** -0.0157*** 0.00233*** (4.900) (4.650) (-150.370) (2.010) MKT 1.00414*** 1.00282*** 0.581*** 0.67*** (1162.230) (135.160) (232.970) (26.370) SMB 0.224*** 0.211*** 0.0646*** -0.522*** (172.650) (16.440) (11.570) (-9.610) HML -0.01260*** 0.0707*** -0.139*** -0.141*** (-9.410) (5.850) (-24.790) (-2.350) MOM 0.0228*** 0.000912 -0.122*** -0.216*** (26.710) (0.120) (-32.330) (-6.130) # Funds 5,731 31 1,178 4 Observations 627,929 4,836 109,220 624 R2 0,854 0,913 0,555 0,739
The OLS regression results of the US ETF sample are presented in the second column of table 6. The sample consists of 31 ETF and 4,826 observations. The constant of the regression is 0.0010 (0.1%), which indicates that the ETF preformed slightly better than the Fama-French
Benchmark. Both regressions have a high explanatory power, given the high R2. This constant
of both regressions will be used to compare the mutual fund performance with the performance of ETFs later on.
Column 3 of table 6 presents the OLS regression results of the EU mutual fund sample. The sample consists of 1,178 funds listed in France, Germany, the United Kingdom,
the Netherlands, Spain and Italy. The UK funds that are listed in GBP were converted to Euro using the EUR/GBP historical spot rate. All regression coefficients are significant at a 1% significance level. The constant of the regression is -0.0157 (-1.57%). This indicates that European mutual funds have underperformed the Fama & French European benchmark by 1.57%. Figure 8 shows the distribution of all individual mutual fund alphas.
Figure 8 – Distribution of individual EU mutual funds alpha
From the total sample no less than 93.77% (1,008 out of 1,075 funds) underperformed the Fama & French benchmark. The best fund outperformed the Fama & French benchmark with 3.6% and the worst fund underperformed the benchmark with -8.84%. These finding are not in line with the results of Otten and Bams (2002), who found that in 4 out of 5 European countries mutual funds (on aggregate) outperform the market. Furthermore, the conclusion of Otten and Schweitzer (2002) that European mutual fund performance better than US mutual funds is not in line with these results. But, this finding was not in line with their expectations: they expected that European mutual fund would perform worse compared to US mutual funds, since the level of competition is lower in Europe. The relative worse performance of my EU mutual fund sample can partially be explained by the average costs of the European mutual funds. Otten and Schweitzer (2002) found that the average costs of their European mutual fund sample is 1.2%, where the average TER in our mutual fund sample is 1.69%. Thereby, my results regarding EU mutual fund performance are in line with another finding
of Otten and Schweitzer (2002) that European mutual funds underperform their relevant benchmarks. They also state that underperformance of mutual funds can be expected in mature and efficient markets.
Next it’s also interesting to see whether there is a pattern in the alpha of the regression. Otten and Schweitzer (2002) for instance found that the performance of their US mutual fund sample dropped during their sample period. Figure 9 presents the alpha of EU mutual funds plotted over time. There doesn’t seem to be a pattern in the alpha. The EU mutual funds underperformed the Fama & French benchmark by 15% at the end of 2008, but they outperformed the benchmark by 10% in the 2009.
Figure 9 – EU mutual fund alpha over time
The last column of table 6 shows the OLS regression results for the EU ETF sample. The sample consists of 4 ETFs and 624 observations. Again all coefficients are significant at a 1% significance level. The constant of this regression is 0.0023 (0.23%), which indicate that the EU ETF sample performed slightly better than the Fama French benchmark that consists of stocks from Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, the Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, and the United Kingdom.
Next I investigated mutual fund performance during the 2008 Global Financial Crisis and the Eurocrisis. For the US, the crisis period is between August 2007 and February 2009. For the European samples the crisis period is between August 2007 and May 2012. The results of the OLS regression are presented below in table 7.
Table 7 – OLS regression results for all samples during the crisis periods
This table presents the OLS estimations of the Carhart (1997) four-factor model for all (MF and ETF) samples for the 2002-2014 period. The dependent variable RET are the monthly returns of the funds/ETFs. MKTRET is the Fama-French market excess return factor (rm – rf). For the US, the
market return is the value-weight return of all CRSP firms incorporated in the US and listed on the NYSE, AMEX of NASDAQ. For the European samples, the market return includes returns of stocks from Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, the Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, and the United Kingdom. The risk-free rate is the 1-month T-Bill rate. SMB is small minus big and HML is high minus low. The construction of SMB and HML follows Fama and French (1993). The momentum factor (MOM) is defined by Carhart (1997) as the momentum factor, and will be used to capture the one-year momentum anomaly of the Fama French (1993) model. Robust t-statistics are reported in parentheses. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.
US Mutual Funds US ETF EU Mutual Funds EU ETF
Coefficients Coefficients Coefficients Coefficients (Robust t-statistic) (Robust t-statistic) (Robust t-statistic) (Robust t-statistic) Constant 0.00151*** 0.00366*** -0.0178*** -0.00166* (15.830) (5.690) (-102.990) (-1.700) MKT 1.0849*** 1.0715*** 0.538*** 0.625*** (576.580 (79.090) (173.820) (37.350) SMB 0.178*** 0.150*** 0.17*** -0.197*** (40.390) (4.880) (22.320) (-4.660) HML -0.143*** 0.0399 -0.0251*** -0.00786 (-41.900) (1.520) (-2.570) (-0.140) MOM 0.0348*** -0.0180 -0.115*** -0.202*** (18.220) (-1.340) (-22.210) (-7.0100) # Funds 4,101 57 995 18 Observations 75,195 1,140 46,710 1,044 R2 0.873 0.907 0.623 0.769
The first column shows the estimates for the US mutual fund sample. This sample consisted of 4,101 funds and 75,195 observations. All coefficients are significant at a 1% significance level. The constant of the regression is 0.00151 (0.15%). This outperformance of 0.15% is
better than the 0.012% outperformance during the 2002-2014 period. This means that compared to the market, mutual funds performance relatively better in the crisis period than in the 2002-2014-sample period. The distributions of all individual mutual fund alphas are shown in figure 10.
Figure 10 - Distribution of individual US mutual funds alpha during the crisis period
Of all funds in the US mutual fund crisis sample with enough data, 31.74% (1,298 out of 4,089) underperformed the Fama & French benchmark. During the 2002-2014 period 45.73% of the funds outperformed the benchmark, so during the crisis relatively more funds have been able to outperform. In the crisis sample there is one fund with an alpha of 13.00%, the alphas of all other funds are lower than 8%. Dropping this best performing fund doesn’t affect the alpha of the regression, so the outperformance of the US mutual funds during the crisis is not only caused by a few good performing funds. These results are in line with the hypotheses that mutual funds have outperformed their benchmark after fees during the crisis period, and that they have performed relatively better during the crisis (compared to the market). The better performance can be a result of the reduction mutual fund holdings during the crisis (Ben-David, Franzoni and Moussawi, 2012). Given the fact that during the crisis the stock prices have decreased most of the time, the reduction of holding can be the main cause of the outperformance during the crisis. Figure 11 in the appendix shows the alpha of the US mutual fund sample during the crisis. There is no clear pattern visible.
The second column presents the results of the US ETF sample during the crisis period. Since there were more ETFs available at the beginning of the crisis (August 2007) compared to 2002, the sample consists of more ETFs: 57, which correspond to 1,140 observations. The constant is 0.00366 (0.37%) and is significant at a 1% significance level. This is higher than the 0.10% of the US ETF sample during the 2002-2014 periods. This means that the ETF sample performed better during the crisis, compared to the Fama French benchmark.
The regression results for the European samples are presented in the last two columns of table 7. For the EU mutual fund sample, the results are shown in column 3. All coefficients are significant at a 1% significance level. The sample consists of 995 funds and 46,710 observations. The constant of the regression is -0.0178 (-1.78%). This performance is worse than the performance of the EU mutual fund sample during the 2002-2014 period (-1.57%). The distributions of all individual mutual fund alphas are shown in figure 12.
Figure 12 - Distribution of individual EU mutual funds alpha during the crisis period
Now 95.23% (938 out of 985) of the EU mutual fund underperformed the Fama & French benchmark during the crisis. During the 2002-2014 period 93.77% of the EU mutual fund underperformed, so during the crisis even more funds underperformed. In the EU mutual fund crisis sample, 3 funds underperformed the Fama & French Benchmark more than 8%. These funds hardly influence the alpha of the regression, so the underperformance is not caused by a few bad performing funds. As for the results during the 2002-2014 sample period for the EU
